The Advanced Television Systems Committee (ATSC) published a Digital Television Standard in 1995 as Document A/53, hereinafter referred to simply as “A/53” for sake of brevity. Annex D of A/53 titled “RF/Transmission Systems Characteristics” is particularly incorporated by reference into this specification. So is Section 5.6.3 titled “Specification of private data services” from Annex C of A/53. In the beginning years of the twenty-first century some in the DTV industry made efforts to provide for more robust transmission of data over broadcast DTV channels without unduly disrupting the operation of so-called “legacy” DTV receivers already in the field.
Samsung Electronics Co., Ltd. proposed introducing convolutionally coded ancillary data into adaptation fields of the 187-byte MPEG-2-compatible data packets included in the 207-byte data segments of the 8VSB DTV broadcast signals used in the United States. This scheme, called “A-VSB”, was championed because the packet decoders in legacy DTV receivers simply disregard the adaptation fields of the 187-byte MPEG-2-compatible data packets containing the convolutionally coded ancillary datastream. This provides a form of backward compatibility in which those legacy DTV receivers can still receive a principal datastream transmitted in the payload fields of the 187-byte MPEG-2-compatible data packets. There is no backward compatibility in the sense that the information content in the convolutionally coded ancillary datastream can be received by legacy DTV receivers. The code rate of A-VSB is one-half the code rate of ordinary 8VSB in its less robust form or one-quarter the code rate of ordinary 8VSB in its more robust form. A-VSB uses a specially designed form of serial concatenated convolutional coding (SCCC) that incorporates the 2/3 trellis coding characteristic of 8VSB DTV signals as its inner convolutional coding. This special form of SCCC is not systematic; that is, the data do not appear in their original form in the signal resulting from the serial concatenated convolutional coding.
A-VSB confines the outer convolutional coding of the SCCC to adaptation fields in 187-byte MPEG-2-compatible data packets used in ordinary 8VSB transmissions, and the adaptation fields are ordinarily constrained to be much less than the 184 bytes the MPEG-2 standard makes available for adaptation field information. Previous proposals made to the ATSC for transmitting robust data have confined those transmissions to the 184 bytes the MPEG-2 standard makes available for payload data. The designs of the transport stream multiplexer for the transmitter is complex in these prior-art proposals, and broadcast studio practice is complicated. If MPEG-2-compatible data packets are used as the primary vehicle for data transmissions, they need to be accompanied by parity information that can ascertain whether the packets have been accurately recovered at the DTV receiver.
The (207, 187) Reed-Solomon forward-error-correction coding specified by A/53 is one way to generate that parity information, and so entire (207, 187) R-S FEC codewords may be robustly coded. It is cumbersome to pack (207, 187) R-S FEC codewords transmitted at 1/2, 1/3, 1/4 or 1/5 ordinary code rate into less than whole 207-byte data segments. However, it is less cumbersome if the number of bytes in the windows is a submultiple of 207, such as nine or twenty-three.
Proponents of robust data transmissions have in the past feared transmitting robust data in all 207 bytes of 8VSB data segments for fear legacy DTV receivers would mistake the data segments for correct or correctable (207, 187) Reed-Solomon codewords. The Reed-Solomon error-correction decoding circuitry in a legacy DTV receiver supposedly would then dispense incorrect 187-byte video data packets to the MPEG-2 decoder and incorrect 187-byte audio data packets to the AC-3 decoder. This is a mischaracterization of the actual legacy receiver problem.
At least eleven of the 207 bytes of a (207, 187) R-S FEC codeword have to disagree with the other bytes for the codeword to be found to contain uncorrectable error. The chance of all eight bits in one byte not being considered to be in error is one in two raised to the eighth power—i.e., one chance in 256. The chance for none of the bits in eleven 8-bit bytes being considered to be in error is one chance in 256 raised to the eleventh power, which is to say one chance in two raised to the eighty-eighth power or one chance in 524 288. So there is one chance in 524 288 that a randomly generated 207-byte segment will be found to be a correct or correctable (207, 187) R-S FEC codeword. (An important point here is that there are 524 287 times more 207-byte data segments that are not correct or correctable R-S FEC codewords than are; this is why the R-S FEC coding is as powerful as it is.) There is a one in two chance the transmitted “TEI bit” will be a ONE rather than a ZERO. This reduces the chance of a randomly generated 207-byte segment being found to be a correct or correctable (207, 187) R-S FEC codeword to one in 1,048,576. If it is still found to be a correct or correctable (207, 187) R-S FEC codeword, the segment will have to include a packet identification (PID) of interest at the time. The probability of this is a small multiple of one in two raised to the thirteen power since the PID has 13 bits. This reduces the possibility of a legacy receiver finding a randomly generated 207-byte segment to be useful to somewhere around one in 4,294,967,296. Once in 6,882,960 data frames. Once every 5736 minutes or so. Once about every four days on average.
The actual concern, then, is not that legacy DTV receivers mistake the data segments for correct or correctable (207, 187) Reed-Solomon codewords. The actual concern is that legacy DTV receivers from one principal manufacturer were designed to evaluate whether DTV reception was acceptably good in response to their Reed-Solomon decoders finding more than a specified number of segments in a data field to be correct or correctable (207, 187) Reed-Solomon codewords. If these particular legacy DTV receivers find fewer than this number of correct or correctable (207, 187) Reed-Solomon codewords in data fields, they infer lack of signal-to-noise ratio high enough for acceptable DTV reception and discontinue normal operation. To accommodate this egregious design error robust transmissions may be confined to 187-byte data segments that are subsequently (207, 187) Reed-Solomon forward-error-correction coded, convolutionally byte interleaved and 2/3 trellis coded.
In DTV receivers specifically designed for receiving robust transmissions, the parity information used to ascertain whether the packets have been accurately recovered at the DTV receiver may be the parity information generated by robustly coding the 187-byte MPEG-2-compatible data packets, rather than being generated by Reed-Solomon coding. It is cumbersome to pack 187-byte MPEG-2-compatible data packets transmitted at 1/2, 1/3, 1/4 or 1/5 ordinary code rate into windows smaller than 187 bytes per data segment. However, it is less cumbersome if the number of bytes in the windows is a submultiple of 187, such as eleven or seventeen.
Parallel concatenated convolutional coding PCCC that reduces code rate to one-third the original code rate is well known in general to the prior art. Such code rate reduction was characteristic of the turbo codes as originally propounded by Berrou, Glavieux and Thitimajshima in their paper “Near Shannon Limit Error-correcting Coding and Decoding: Turbo-codes” published in the 1993 Proceedings of IEEE International Communications Conference. The Universal Mobile Telecommunications System (UMTS), one of the two most widely adopted third-generation cellular standards, employs turbo coding that reduces code rate by a factor close to three. These turbo codes each comprise three parts similar to each other in size: (1) the original data, (2) parity information developed by first convolutional coding of the original data and (3) further parity information developed by second convolutional coding of the original data. In these turbo codes the original data is interleaved differently during the second convolutional coding than during the first convolutional coding, with the respective interleaving preferably being as random as possible. The other most widely adopted third-generation cellular standard cdma2000 uses different interleaving than UMTS and employs turbo coding that reduces code rate by a factor close to five. Punctured forms of cdma2000 reduce code rate by factors of two, three or four.
The inventor observed that PCCC is more robust than simple outer convolutional coding. This observation enabled him to discern that serially concatenating PCCC with 2/3 trellis coding should provide a more robust form of 8VSB DTV broadcasting likely to be preferable to SCCC that serially concatenates simple outer convolutional coding with 2/3 trellis coding as inner convolutional coding. Field testing of A-VSB has subsequently confirmed that reception of the half-code-rate format is substantially inferior to reception of the quarter-code-rate format. This undermines a basic reason for considering SCCC that serially concatenates simple outer convolutional coding with 2/3 trellis coding as inner convolutional coding—namely, lower reduction in code rate than possible with PCCC that does not use puncturing. PCCC that reduces code rate by a factor of three will be preferable to the outer convolutional coding of A-VSB that reduces code rate by a factor of four, if reception is at least almost as good. When PCC is considered by itself, it is known that PCCC that reduces code rate by a factor of three brings one close to Shannon limit and further reductions of code rate provide smaller improvements in approaching that limit. In view of this known fact the inventor has guessed that PCCC that reduces code rate by a factor of three will, when concatenated with 2/3 trellis coding, secure performance in an AWGN channel that is close enough to optimal not to justify further reduction in code rate.
The inventor discerned that systematic turbo coding that reduces code rate to a fraction of the original code rate while retaining the original form of data packets is of especial interest to 8VSB DTV broadcasting. Such systematic turbo coding can provide the broadcaster with the option of transmitting 187-byte MPEG-2-compatible data packets of original data in (207, 187) R-S FEC codewords that can be usefully received by legacy DTV receivers. If code rate is reduced to one-third the original code rate, for example, these (207, 187) R-S FEC codewords can occupy 104 of the 312 segments of a data field with the remaining 208 data segments being occupied by parity information for the turbo coding. Two standard-definition digital television (SDTV) signals can be continuously robustly transmitted in this manner without disenfranchising legacy receivers. Insofar as theoretical throughput capability is concerned, this compares favorably with A-VSB robustly transmitting the same DTV signal at one-half ordinary code rate and is also contemporaneously transmitted by ordinary 8VSB at full code rate. The turbo coding that reduces code rate by three is financially advantageous when the same copyrighted information is transmitted robustly as well as to legacy receivers. This is because copyrighted information is transmitted once by the broadcaster, rather than twice, which may reduce the copyright royalties to be paid by the broadcaster. Even when copyright royalties are not in issue, the turbo coding that reduces code rate by three is financially advantageous when the same information is transmitted robustly as well as to legacy receivers. This is because broadcasters are required to pay spectrum usage fees for ancillary transmissions that can in no part be usefully received by legacy DTV receivers.
In 2007 Samsung engineers proposed adapting their A-VSB transmission system for mobile reception by DTV receivers that are carried by fast-moving vehicles such as automobiles, buses or railroad passenger cars. Such reception is disrupted by momentary “deep fades” or drop-outs in received signal strength as the vehicle moves through underpasses or passes large buildings blocking the transmission path. To help a mobile DTV receiver withstand these momentary drop-outs, the Samsung engineers introduced an outer byte interleaver after the encoder used to generate the (207, 187) R-S FEC codewords supplied for serial concatenated convolutional coding. This outer byte interleaver spread the successive bytes of each (207, 187) R-S FEC codeword apart so far that fewer of them would be lost during a momentary drop-out. Hopefully, so few bytes would be lost in each (207, 187) R-S FEC codeword that the Reed-Solomon decoding apparatus in a DTV receiver designed for mobile reception would be able to correct the codeword and restore the missing bytes.
The inventor perceived that the Samsung proposal had a basic flaw in regard to a DTV receiver designed for mobile reception of turbo coded DTV signals. Turbo coding primarily benefits the AWGN performance of a receiver. Much of the improvement in the AWGN performance of a receiver at low signal-to-noise ratios (SNR) derives from iteration of the turbo decoding procedures. Reed-Solomon forward-error-correction coding the data to be turbo coded has been considered previously for improving the ability of the receiver to withstand burst error as well. However, the inventor perceived that R-S FEC coding has a further advantage in that R-S decoding can be used to determine when iteration of turbo decoding procedures allows the R-S decoding to restore the data in the transmitted R-S FEC codeword to its original form. The iteration of the turbo decoding procedures can be discontinued for that data, thus to conserve the power that would otherwise be consumed by further iteration. Keeping power consumption of the DTV receiver low is a major design concern, especially for receivers to be sold in California, which has stringent limitations on power consumption in consumer devices.
The outer byte interleaver that Samsung engineers introduced between R-S FEC coding and turbo coding in the DTV transmitter requires a matching outer byte de-interleaver between turbo decoding and outer R-S decoding in the DTV receiver. This outer byte de-interleaver has such long latent delay associated with it that feedback from the outer R-S decoding is generated too late for timely shutting down iteration of turbo decoding procedures. The inventor concluded that the outer byte interleaving in the DTV transmitter should be done after both R-S FEC coding and turbo coding are completed.
The inventor reasoned that this would permit re-positioning the outer byte de-interleaver to precede the cascade connection of turbo and outer R-S decoders in a DTV receiver, supposing that outer R-S decoder was to feed back control information to the turbo decoder. This placement does not facilitate the 2/3 trellis decoder being included within the turbo decoding apparatus as Samsung engineers do in A-VSB, however. The 2/3 trellis decoding has to be considered to be a preliminary decoding procedure that is serially concatenated with the subsequent turbo decoding procedure. This observation led the inventor to contemplate outer de-interleaving being done after 2/3 trellis decoding, convolutional byte de-interleaving, decoding of (207,187) R-S FEC coding and data de-randomization were performed as prescribed by A/53. Turbo decoding would then follow the outer de-interleaving. A convenient feature of this arrangement is that de-randomization is completed before iterative turbo decoding and is done at a point in the system where the time base is well defined. The soft-decision information associated with the preliminary 2/3 trellis decoding procedure can be passed along to the subsequent turbo decoding procedure even though there are intervening steps of data randomization, de-interleaving, and possibly symbol re-coding. The turbo coding can use parallel concatenated turbo code, already proven in wireless communications, which would constrain code rate to being no greater than one-third that of ordinary 8VSB.
In the A-VSB system the tail bits of the turbo code encoding procedures are discarded. Preserving the tail bits of the turbo code encoding procedures improves decoding performance at low SNR by facilitating sweeps through the trellis in reverse direction as well as forward direction. This tends to reduce the number of iterations required for correcting bit errors, so decoding can proceed at slower speeds and consume less power. This also facilitates forward-backward turbo decoding algorithms correcting bit errors in both initial and final portions of a turbo codeword that are separated by burst error resulting from a protracted deep fade. Correcting the bit errors caused by AWGN in the initial and final portions of a turbo codeword thus corrupted increases the chances of being able to correct the recovered data by subsequent R-S decoding procedures. That is, the forward-backward turbo decoding algorithms can work from both ends of a turbo codeword to attempt to close the extent of a burst error that the subsequent R-S decoding procedure must then attempt to overcome.
Part of the A-VSB system proposed by Samsung engineers concerns supplemental training signals being interspersed throughout transmitted fields of trellis-coded data. These supplemental training signals are introduced as private data within adaptation fields of successive MPEG-2 compatible packets, including those packets that are involved in ordinary 8VSB transmissions as well as the null packets that are employed for A-VSB turbo code transmissions. These supplemental training signals were touted as being necessary for adaptive equalization being able to track the fast-changing dynamic multipath reception conditions encountered in mobile reception. Most current DTV receiver designs use forms of adaptive equalization that rely on incremental auto-regression techniques of one sort or other. They usually combine elements of the Wiener technique, which relies on training signal for developing error signals to adjust equalization, and the Kalman technique, which develops error signals to adjust equalization from ordinary data symbols. Because these adaptive equalization techniques use incremental feedback adjustments, they are inherently slow, Kalman-type auto-regression techniques generally being slower than the Wiener-type auto-regression techniques that rely on training signal. Supplemental training signals support somewhat faster tracking of rapidly changing multipath reception conditions in adaptive equalization techniques that use incremental feedback adjustments.
In another technique, attributable to Dr. J. Douglas McDonald, the channel impulse response (CIR) is computed by auto-correlating a sliding window 4096 symbols wide. Computation of the CIR is done using discrete Fourier transform, or DFT. Incidentally, this facilitates initial frequency-domain equalization in a DTV receiver to whiten the channel frequency response. Each computation of CIR can be made very quickly. If multipath conditions are slow-changing, the CIR computations can be averaged over longer time periods to improve accuracy of the equalization. If multipath conditions change rapidly, however, the averaging can be curtailed to speed up equalization adjustment much more than possible with Kalman-type or Wiener-type auto-regression techniques or variants of those techniques. The bottom line is that the data-randomized 8VSB signal contains enough information to implement adaptive equalization even when multipath conditions change rapidly without need for additional training signals.
Accordingly, the transmission of turbo coding in the adaptation fields of MPEG-2-compatible null packets following several bytes of supplemental training signal per A-VSB unnecessarily costs code overhead. A better procedure is to pack turbo coding into the entire 184-byte payload-data fields of MPEG-2-compatible null packets. A special PID could be established to identify the packets having their payload-data fields packed with turbo coding, of course, rather than null packets being used.
The inventor initially considered turbo coding (207, 187) R-S FEC codewords individually, so the internal bit interleaver in the turbo code encoder would have to permute 207×8=1656 bits in each resulting data segment. However, he considered whether higher interleaver gains could be fitted into the 8VSB signal. The inventor observed that eight (207, 187) R-S FEC codewords fit exactly into nine 184-byte payload-data fields, which suggested to him that the input signal to the turbo code encoder could conveniently be parsed into blocks of 207×8×8=184×9×8=13,248 bits. This sets a size for the internal bit interleaver that is within the capability of bit interleaver designs currently used in turbo code encoders for wireless telephony. Presuming the turbo code encoder to use two recursive systematic convolutional code encoders that are parallel concatenated, each block of input signal consisting of 13,248 bits generates a respective turbo codeword consisting of 3×13,248+12=39,756 bits. The twelve additional bits are turbo code tail bits.
In A-VSB the tail bits associated with turbo codewords are discarded before packing into the adaptation fields of MPEG-2-compatible data packets. It is preferable to transmit the tail bits associated with turbo codewords, however. The tail bits improve forward and back decoding procedures for turbo codewords, particularly when reverse sweeping through the trellises. The improved decoding of each turbo codeword tends to reduce the number of times the forward and back decoding procedures need be iterated to obtain satisfactory bit error rate for low-SNR AWGN reception conditions. This furthers the primary objective of the invention to reduce the power consumed by the receiver during decoding of turbo code.
Furthermore, improving the capability to reverse sweep through the trellises helps to overcome drop-outs in signal strength that occur during the mid portions of turbo codewords. The portion of the turbo codeword occurring after the deep fade is more quickly decoded. Accordingly, a sufficient amount of the R-S FEC codeword to permit its correction is likely to be earlier available. The correction of the R-S FEC codeword restores the data lost because of the deep fade.
The inventor discerned that the successive bytes of turbo codewords should be transversely disposed relative to the payload-data fields of the MPEG-2 compatible packets they are packed into before being time-division multiplexed with other MPEG-2 compatible packets to form data fields. The MPEG-2 compatible packet in these data fields are provided with inner (207, 187) R-S FEC coding and subjected to inner byte interleaving, with the resulting data then being encoded with 12-phase 2/3 trellis code. The inner byte interleaving is of a convolutional type that spreads the bytes in each inner (207, 187) R-S FEC codeword to be 52 byte intervals apart. Because the inner byte interleaving is of convolutional type, there are periodic snaps back in time of 51 data segment intervals.
If the bytes of turbo codewords are interleaved correctly, then each byte of a turbo codeword occupies the same position within the payload-data field of a respective MPEG-2 compatible packet and the inner (207, 187) R-S FEC codeword generated therefrom as the other bytes of that turbo codeword. So, the inner byte interleaving shifts all the bytes of each turbo codeword by the same amount in time, and the intervals between successive block-interleaved bytes remain alike. Accordingly, the convolutional inner byte interleaving does not affect a turbo-coded outer R-S FEC codeword, the bytes of which are block-interleaved, so as to alter the capability of that codeword to overcome a deep fade.
Positioning codewords transversely across the payload-data fields of MPEG-2 compatible packets transmitted by 8VSB has other advantages. There is no need to constrain the length(s) of codewords, so that each will fit exactly within the 184-byte width of a respective payload-data field, or so that a whole number of codewords will fit exactly within a reasonably small multiple of that 184-byte width. Also, there need be no concern that coding artifacts will appear to be echoes to the adaptive equalization filtering in DTV receivers. There are fewer, if any, constraints as to how data segments containing other 8VSB signals are time-division multiplexed with the data segments containing turbo coding. The inventor noted that transverse interleaving by the outer byte interleaver results in the inner and outer R-S coding being cross-interleaved, and so essentially comprising cross-interleaved Reed-Solomon codes (CIRC). This holds out the possibility that CIRC techniques might eventually be found to be of use in difficult decoding situations.
The inventor spent some time seeking a technique to construct outer interleavers that would position codewords transversely across the payload-data fields of MPEG-2 compatible packets and could also overcome deep fades lasting as long as a second. Convolutional interleaver designs that he initially attempted required very large numbers of temporary storage locations in memory, leading him to consider block interleaver designs and combinations of convolutional and block interleaver designs. None of these designs were completely satisfactory, so he conducted a thorough search of patents to interleavers. He found a type of block interleaver described generally in U.S. Pat. No. 5,907,560 issued 25 May 1999 to P. M. P. Spruyt and titled “Method for interleaving data frames, forward error correcting device and modulator including such a device” that could be adapted to provide the desired outer byte interleavers. The required number of temporary storage locations in memory were substantially lower than in the byte interleaver design inspired by the Spruyt patent than other interleaver designs the inventor had considered. Analysis of the interleaver problem working back from knowledge of these various interleaver designs seemed to indicate that the number of temporary byte-storage locations in memory could not be further reduced.
During the course of his work the inventor discerned that the placement of bytes of the turbo coding within the turbo codewords supplied for outer byte interleaving was crucial in securing best performance of the R-S FEC coding in overcoming deep fades. Bytes of the parity bits associated with each byte of data should be closely grouped with that byte of data in the turbo codeword supplied for outer byte interleaving by Spruyt's method. This minimizes the amount of the turbo codeword that a deep fade renders unfit for turbo decoding. More of the turbo codeword survives for turbo decoding. Turbo decoding in the forward direction from the beginning of the codeword can recover more data from the initial surviving portion of the codeword than otherwise possible. Turbo decoding in the reverse direction from the conclusion of the codeword can recover more data from the final surviving portion of the codeword than otherwise possible. Accordingly, the hiatus in data that decoding of the R-S FEC coding must correct for is kept as small as possible.
Another known technique for overcoming fading is called “staggercasting”, a variant of which Thomson, Inc. has proposed be used in robust 8VSB transmissions. Staggercasting communications systems transmit a composite signal including two component content-representative signals, one of which is delayed with respect to the other. The composite signal is broadcast to one or more receivers through a communications channel. At a receiver, delayed response to the earlier transmitted component content-representative signal supplied from a buffer memory is contemporaneous in time with the later transmitted component content-representative signal. Under normal conditions, the receiver detects and reproduces the content of the later transmitted signal as soon as it is received. However, if a deep fade occurs, then the receiver detects and reproduces the content of the earlier transmitted signal as read from buffer memory. If the delay period and the associated delay buffer are large enough, then fairly long deep fades can be overcome. This capability not only requires a several fold increase in the amount of memory required in a receiver; it halves the effective code rate of the transmission.
The inventor perceived that the processing of soft decisions in turbo decoding allows a more sophisticated approach to be taken for the reception of staggercasting. Soft decisions concerning the contents of an earlier transmitted turbo codeword and concerning the contents of a later transmitted repeat of the earlier transmitted turbo codeword can be analyzed for selecting which of corresponding portions of the two turbo codewords as received is more likely to be correct. The selection procedure can synthesize a turbo codeword that is more likely to be correct than either of the turbo codewords from which the parts of the synthesized turbo codeword are drawn. The synthesized turbo codeword can then be subjected to turbo decoding and R-S decoding procedures.
The inventor discerned that this synthesis procedure can provide more than a tenfold increase in the capability of the turbo coding to withstand dropouts in received signal strength with only a doubling of receiver memory. This is accomplished by using a novel form of iterative-diversity transmissions or “staggercasting” in which each successive turbo codeword is immediately repeated in its transmission.